Submitted for publication. Updated on September 4, 2014.
Restructuring the Backhoe Loader Product Line at
Caterpillar: A New Lane Strategy
Masha Shunko
Krannert School of Management, Purdue University, mshunko@purdue.edu
Tallys Yunes
School of Business Administration, University of Miami, tallys@miami.edu
Giulio Fenu
Caterpillar Inc., fenu giulio@cat.com
Alan Scheller-Wolf
Tepper School of Business, Carnegie Mellon University, awolf@andrew.cmu.edu
Valerie Tardif
SAP, vtardif@sap.com
Sridhar Tayur
Tepper School of Business, Carnegie Mellon University, stayur@andrew.cmu.edu
Caterpillar recently embarked on an ambitious program to radically change how it markets and sells key
products of its Building Construction Products division (BCP). The goal was to move from a strict build-toorder strategy, in which customers selected one of millions of possible configurations, to a Lane Strategy in
which the majority of their customers would choose machines from just over 100 configurations. To successfully make such a radical change Caterpillar needed to quantify how customers would potentially react to the
new strategy, and how such a drastic simplification of their product line would affect their manufacturing,
sales, and service costs. We embarked on a study with Caterpillar to explicitly model customers’ reactions to
reduced product lines, to estimate the (positive and negative) effect such variety has on Caterpillar’s costs
– the cost of complexity – and ultimately help them design and implement this strategy for their flagship
BCP product, the Backhoe Loader. Based on our analysis, Caterpillar began implementing the new strategy
with their 2010 price list, moving completely to the new strategy in 2011. Since that time Caterpillar has
expanded their Lane strategy throughout all of their product lines, fundamentally remaking their business.
Key words : product portfolio optimization; cost of complexity; manufacturing; machinery
1. Introduction
In 2010 Caterpillar (CAT) unveiled a dramatically new strategy for pricing and marketing their
“BHL” series of small backhoe loaders, one of the most popular products within their Building
Construction Products (BCP) division. This new strategy has radically changed how BCP markets
and sells their small machines, focusing the bulk of CAT’s customers on a few popular models.
1
2
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Previously, BCP offered customers an almost unlimited variety of products, built-to-order, priced
according to an itemized price list. This maintained very high customer satisfaction, but greatly
complicated BCP’s supply chain and service operations: With such broad demand dealers had to
hold large amounts of inventory to give a representation of the many possible choices. Moreover,
Caterpillar had to maintain documentation and provide service for an extremely heterogeneous
group of machines, driving up their costs. Finally, as demand was fragmented across thousands of
configurations, forecasting was problematic, leading to frustration among CAT’s supply base.
Thus CAT’s BCP division saw great potential for a rationalization of their BHL product line.
But there were three crucial questions that needed to be answered before such a new strategy could
be devised and implemented:
1. “How would customers react to a product line reduction?” Answering this question required
developing an understanding of how customers value different machines.
2. “How much could be saved by focusing their BHL product lines?” Answering this question
required developing an understanding of the general form of the cost of complexity.
The answers to these two questions could help Caterpillar answer the ultimate question:
3. “How should we configure our new BHL product line?” Specifically, which machines should
we offer, and at what prices?
The primary contribution of this paper is to demonstrate the power of our three-step analytical
framework for product line simplification. Step 1 answers question 1 by capturing customer behavior
using migration lists; step 2 answers question 2 by creating a detailed mathematical representation
of the company’s cost of complexity (CoC); and step 3 answers the final question by combining the
migration lists and the CoC function into an optimization model that proposes an improved product
line. The generality and flexibility of our framework stem from the fact that the mathematical
and statistical techniques used in steps 1, 2, and 3 can be tailored to the situation at hand, as
long as they produce the output required by each subsequent step. We also demonstrate that our
framework can be used as an effective what-if tool for managers, allowing them to successfully
evaluate different solutions under varying problem conditions.
To answer CAT’s first question we leveraged BCP’s extensive dealer network to gain an understanding of the segmentation, preference patterns, and price sensitivities of BCP’s customer base.
We combined this dealer knowledge with the entire line’s sales history over the previous two years
to construct a detailed analytical model of customer preferences and substitution (see Section 4).
This produced a model of how BCP’s customers would react to changes in the product line.
To answer the second question, we built a detailed model to estimate the total direct and indirect
costs of complexity, using an extensive empirical analysis of CAT’s cost data, as well as surveys with
Caterpillar engineering and marketing experts. This model captured both variety-based (driven by
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
3
the number of options offered) and attribute-based (driven by specific complex options) costs and
benefits of product diversity, from manufacturing, to spare parts, to sales effort. (See Section 5).
We then combined the customer and cost models within a mathematical programming model,
in Section 6, to evaluate different product lines against randomized demand patterns and market
scenarios. In coordination with expert input from CAT, this step determined the right product
mix for the line, offering customers broad choice while also controlling the costs of complexity.
The outcome of the project was implemented as a new Lane strategy, offering machines within
three different lanes: Lane 1, the Express Lane, featuring four built-to-stock configuration choices
at an expected lead time of a few days; Lane 2, the Standard Lane, featuring 120 predefined
configurations, built-to-order at an expected lead time of a few weeks; and Lane 3, the A-La-Carte
Lane, built to order machines with an expected lead time of a few months.
Caterpillar committed to a phased roll-out of the project, publishing both an (old) a-la-carte
price list and a (new) Lane price list in 2010, and transitioned to a single Lane price list in 2011.
The Lane 1 configurations were immediately able to capture a significant portion of demand,
contributing to a reduction in warranty costs on the order of 10%, as predicted by our analysis.
Caterpillar has continued expanding and refining their BHL lane strategy – for example, they have
now reduced the Lane 1 configurations to only two. In addition, Caterpillar has applied variations of
our cost of complexity analysis to other divisions within the firm, helping to guide their extensions
of the lane approach – a fundamental strategic change – to the entire company.
To the best of our knowledge, no previous work has ever combined an empirically-developed
CoC function as detailed and comprehensive as ours, with customer preferences regarding product
substitution, in an optimization algorithm that was implemented with real-life data, at an industrial
scale. Moreover, our work ultimately produced recommendations that were actually implemented
and verified to generate significant improvements.
In Section 2 we place our work within the product line optimization and practical application
literature. Then, in Section 3 we briefly describe the BHL lines. In Section 4 we present our key
tool for answering the first of Caterpillar’s questions (“How would customers react to a product
line reduction?”), our customer migration list model. Next, in Section 5, we describe how we
answered the second question (“How much could be saved by focusing their BHL product lines?”)
by presenting our Cost of Complexity model. We then combine these in our optimization model, in
Section 6, to answer the ultimate question, (“How should we configure our new product line?”). We
present the results and insights from our analysis in Sections 7 and 8, and conclude in Section 9.
2. Literature Review
The marketing literature illustrates the difficulty of predicting the effects of reducing product line
complexity: Some works describe how narrowing a product line may detract from brand image or
4
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
market share, e.g. Chong et al. (1998), while others posit that reducing the breadth of lines and
focusing on customer “favorites” may actually increase sales, see for example Broniarczyk et al.
(1998). Our model is consistent with both of these streams: If a customer finds a product that
meets her needs (i.e. a “favorite”) she will make a purchase; if such a product and its acceptable
alternatives are no longer part of the product line, she will not.
There is also a long history of empirically studying the impact of product line complexity on costs.
Foster and Gupta (1990) analyze the responses to a questionnaire to assess the impacts of volumebased, efficiency-based, and complexity-based cost drivers within an electronics manufacturing
company. They find that manufacturing overhead is associated with volume, but not complexity or
variety. Banker et al. (1995) use data from 32 manufacturing plants to evaluate the same question;
they find an association of overhead costs with both volume and transactions, which they take
as a measure of complexity. Anderson (1995) examines the effect of product mix heterogeneity on
overhead costs in three textile factories. Of the seven different types of heterogeneity she identifies,
two are shown to be associated with higher overhead costs. Fisher and Ittner (1999) analyze data
from a GM assembly plant, finding that option variety contributes to higher labor and overhead
costs. We complement these works by explicitly formulating and calibrating a detailed model
to estimate the total direct and indirect costs (and benefits) of complexity for the BHL line at
Caterpillar, based on expert surveys and empirical analysis of cost data.
Product line optimization has a rich literature: Kok et al. (2009) and Tang (2010) provide recent
surveys. Several recent papers consider the strategic selection of a product line via equilibrium
analysis: Alptekinoglu and Corbett (2008), Chen et al. (2008), Chen et al. (2010), and Tang and
Yin (2010). Our paper uses math programming to optimize a more detailed model.
Bitran and Ferrer (2007) consider the problem of determining the optimal price and composition
of a single bundle of items and a single segment of customers in a competitive market. They provide
extensions to multiple segments or multiple bundles based on mathematical programming, but
this latter problem becomes very complex, and is left as future research. Wang et al. (2009) use
branch-and-price to solve the problem of selecting a line to maximize the share of market, testing
their algorithm on problems with a small number of items but many levels of product attributes
on simulated and commercial data. And Schoen (2010) extends the work of Chen and Hausman
(2000) to allow more general costs and heterogeneous customers. None of these algorithms have
been shown to be suitable for problems anywhere near the size and complexity of Caterpillar’s
(thousands of customers and millions of potential configurations). This has led to the investigation
of heuristic line optimization methods: For example Fruchter et al. (2006) apply a genetic algorithm
to simulated and survey data; and Belloni et al. (2008) compare several heuristics to the optimal
solution for small (up to 5-product) instances. Neither of these are actual implementations.
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
5
Kok and Fisher (2007) develop and apply a methodology to estimate demand and substitution
patterns for a Dutch supermarket chain, based on empirical demand data. They develop an iterative heuristic that determines the facings allocated to different categories, and the inventory of
individual elements within the categories. In contrast: (i) We develop an empirical cost of complexity function; (ii) We use a more comprehensive substitution mechanism, the migration list. In
Kok and Fisher (2007), customers who find their first choice absent will substitute at most once
(so if their second choice is absent they leave); (iii) We develop a single product lane strategy for
use by all dealers in the network; and (iv) Our results are based on actual implementation.
Fisher and Vaidyanathan (2011) explore how to select assortments for retail stores; they describe
their work as enhancing a localized choice model to make it operational in practice. Our models
share a localized choice model with randomization, location at extant configurations, and preference
sets for substitution (i.e. our migration lists). But whereas in Fisher and Vaidyanathan (2011) all
customers who prefer a particular product have the same preference set, we randomize the option
utilities of each customer, so customers who purchased the same product may spawn different
migration lists. Furthermore, we use an additive model of attribute utilities, theirs is multiplicative.
Other important differences include: Fisher and Vaidyanathan (2011) estimate demand intensity
and substitution parameters from historical data, whereas we use expert opinion to get utilities,
and generate demand by randomizing past sales. In contrast to our approach that seeks to maximize profits using our empirical cost of complexity function, they maximize revenue with greedy
heuristics. Finally, they show just two examples — snack cakes and tires — implementing a small
set of their recommended changes with the tires line, increasing revenue by 5.8%.
Ward et al. (2010) is another recent paper that rigorously applies an analytical framework to
the product line problem, developing two tools to help Hewlett-Packard. Like Caterpillar, HP has
product lines that could, in theory, span millions of different configurations.
The first tool develops a comprehensive complexity cost function, comprised of variable and fixed
costs, to be used when evaluating the introduction of new products. This complexity cost function
has some similarities to ours, but focuses more on inventory costs, lacking anything related to
our attribute based costing. Furthermore, any substitution effects on inventory, which they term
cannibalization, are subjectively estimated at a high level.
Their second tool uses a heuristic to construct a line from a selection of extant products. This
tool does not use their complexity cost function, nor does it consider substitution — rather it
constructs a Pareto frontier of those top k products that would cover the desired percentage of
historical order demand (or order revenue). So while they seek the appropriate line to satisfy
possibly multi-product orders assuming customers will not substitute, we find the correct line of
products to satisfy orders for individual products in which customers may substitute.
6
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Rash and Kempf (2012) recently published work on product line design and scheduling at Intel.
They find the set of products to produce, for different markets, to maximize profit over a time
horizon while obeying budget and availability constraints. They perform hierarchical decomposition, utilizing genetic algorithms along with MIPs. This work is quite different from ours in that
their demand is viewed as deterministic, so substitution is not included in the model.
The three-step framework we use was first introduced in Yunes et al. (2007), which describes a
product line simplification effort implemented at John Deere & Co. Our current work shares some
of its methodology with Yunes et al. (2007), but extends this work in several dimensions.
Specifically, we: (i) Explicitly calculate and validate estimates of the parts utilities and complexity
function; they were exogenous in Yunes et al. (2007). (ii) Create a sophisticated, endogenous, cost
of complexity function; the function used in Yunes et al. (2007) was exogenous. (iii) Owing to the
form of our endogenous function, we utilize a different optimization procedure, the “differential
approach.” (iv) To achieve the aggressive product line goals of Caterpillar, we make decisions at
the option level, rather than the machine level, as in Yunes et al. (2007). We also incorporate
pricing decisions and migration across models, which are not possible in Yunes et al. (2007).
3. BHL Product Families
Our product line simplification effort at CAT involved four models in the backhoe loader (BHL)
family: 416E, 420E, 430E, and 450E. The 416E is the basic model, while the 420E, 430E and 450E
provide progressively superior horsepower and capabilities. We refer to a complete machine as a
configuration. Each configuration is composed of features; for each feature, a configuration specifies
one of the options within that feature. For example, the feature stick has the options standard
and electronic. Table 1 contains a summary of the features present in each BHL model that were
included in our project, together with the number of options in each. A dash “-” indicates that a
feature is not present in a model or was not included in our analysis.
To create a complete configuration, a customer selects one option for each of its features, while
making sure that these options are compatible. One of the difficulties faced by CAT is that the
number of potential configurations is immensely large. For example, for model 416E in its most basic
version, the number of feasible configurations is 37,920. When we include choices for attachments,
there are 2,275,200 distinct feasible configurations. The vast majority of these configurations have
never been, and most likely will never be, built. The mere fact that they could be purchased,
however, creates overhead costs for CAT’s operations (see Section 5).
So how many configurations are built? Figure 1 depicts the number of different configurations
(left panel) and options (right panel) required to capture specified revenue and sales targets for
eight month’s worth of sales data for model 420E. Of the 569 built configurations, 400 were needed
7
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Table 1
Number of options in each feature of CAT’s BHL
models.
BHL Models
416E 420E 430E 450E
Sticks
2
2
2
2
3
3
3
6
Backhoe Hydraulics
Backhoe Controls
2
5
13
13
4
Loader Buckets
Loader Hydraulics
2
2
2
2
5
5
4
2
Cab/Canopy
Powertrain
4
3
2
2
2
2
Engine Cooling
Counterweights
4
4
4
3
3
3
3
Backhoe Aux Lines
Engine Coolant Heater
2
2
2
2
Product Link
2
2
2
2
2
2
2
Ride Control
Front Loader Mechanics
2
2
Features
Figure 1
Number of distinct configurations (left) and options (right) required to capture given percentages of
revenue and sales volume for BHL model 420E.
100
100
90
90
Revenue
Sales
80
80
70
70
Percentage of coverage
Percentage of coverage
Revenue
Sales
60
50
40
60
50
40
30
30
20
20
10
10
0
0
0
100
200
300
Number of configurations
400
500
600
0
5
10
15
20
25
Number of options
30
35
40
to capture about 95% of revenues and sales volume. Similarly, 41 out of the 45 available options
were needed to capture at least 90% of revenues and sales volume. Therefore, in order to achieve the
sought reductions in product offerings, it was imperative to steer purchases toward a considerably
smaller subset of products and options.
4. Modeling Customer Behavior
The key to evaluating the potential pitfalls of reducing a product line is a good understanding of
customers’ purchasing flexibility: while customers will require that the product they are buying
satisfy some minimum requirements, not every feature needs to be in perfect alignment with their
expectations. In addition, customers typically display some degree of price flexibility.
45
8
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
The centerpiece of our approach to capture flexibility is the migration list, an ordered list of
products within the customer’s price and utility tolerance (see Yunes et al. 2007 for details). The
first configuration on the list is the customer’s first choice; if available the customer will buy that
configuration. If that configuration is unavailable and there is a second configuration on the list
the customer will buy that if available, and so on. If none of the configurations on a customer’s list
are available, that customer buys nothing (i.e. goes to a competitor). As mentioned in Section 2,
this is an enhanced localized choice model, in the spirit of Fisher and Vaidyanathan (2011).
For each unique purchase in our database, we construct a corresponding migration list. Many
factors come into play when constructing a customer’s migration list. Below is pseudo-code showing
how the process works; each step is described in detail in a subsection below.
Construction of Migration Lists: For each customer C who bought a configuration over the past
H months repeat:
1. Let MC be the configuration (machine) bought by C.
2. Apply segmentation rules to MC (Section 4.1) to place C in a segment SC .
3. Based on the price and utility of MC , and on characteristics of segment SC , construct a
randomized list of configurations, LC , as acceptable alternatives to MC (Section 4.2).
4. Sort LC in non-increasing order of configuration utility, pruning it if it exceeds the maximum
allowed length (Section 4.3).
4.1. Customer Segmentation
Customer segmentation plays an important role in the creation of migration lists because it affects
customer flexibility. For example, customers who live in extreme weather conditions are unlikely
to buy a configuration that does not include a cab with climate control, and customers who need
to carry very heavy loads are not willing to sacrifice horsepower. We used focus groups composed
of CAT experts and actual customers to identify the main customer segments and their characteristics. This produced six segments: performance extreme (PE), performance extreme versatility
(PEV), performance mild (PM), performance mild versatility (PMV), commodity extreme (CE),
and commodity mild (CM). The performance category represents customers who are less price sensitive and need powerful machines. The extreme and mild categories refer to weather conditions,
and the versatility category represents customers who need their machines to perform a variety of
tasks. Based on historical sales data, the fraction of customers in each of the above six segments
are approximately 20, 20, 25, 10, 5, and 20 percent.
A set of segmentation rules was created to classify each purchase: given a configuration, its
customer segment is determined by the presence and/or absence of certain options, represented as
part numbers. For example, there are eight ways for a 416E loader to be placed in segment PE.
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Table 2
9
AHP example: Pairwise and Normalized comparisons.
Pairwise Comparisons
Sticks Powertrain Hydraulics
Sticks
1
1
3
Powertrain
1
1
2
1
1
Hydraulics
1
3
2
7
5
Sum
6
3
2
Normalized Comparisons
Sticks Powertrain Hydraulics Avg. Scores
2
1
3
0.4429
7
5
2
3
2
1
0.3873
7
5
3
1
1
1
0.1698
7
5
6
One is: two out of the options 2146913, 2099929, and 2139293 must be present (89HP powertrain
and e-stick), and one out of the options 2044161, 2044162, and 2284602 must be present (cabs),
and the option 2120206 cannot be present (6-function hydraulics), and neither option 2497912,
nor option 2624213 can be present (one-way and combined auxiliary lines).
In addition to the segment-specific option utilities that will be discussed in Section 4.2, the
other segment-specific parameters that affect migration list generation are the reservation price
and reservation utility (see Section 4.3).
4.2. Estimating Utilities
For each of the customer segments identified in Section 4.1, we calculate option utilities as follows.
First, to estimate the importance of a model’s features, we asked a group of CAT employees with
sales and manufacturing expertise to use the Analytic Hierarchy Process (AHP) (Saaty 1980).
AHP asks experts to estimate different feature’s relative importance, on a scale from 1 (equally
important) to 9 (much more important). After these pairwise relationships have been established,
the scores are transformed into absolute scores, the relative importance of each feature. The same
group of experts is then asked to rank the options within each feature on a scale from 0 to 100.
These option scores are scaled so that the option receiving a score of 100 is assigned a value equal
to its feature’s relative importance. These scaled scores represent the option utilities.
We demonstrate this on a simplified example with three features: sticks, powertrain, and backhoe
hydraulics. Assume the experts have ranked the pairwise importance as shown in the “Pairwise”
portion of Table 2. The numbers in the “Sticks” row mean that sticks are as important as powertrain, and three times more important than hydraulics. Similarly, the second row indicates that
powertrain is twice as important as hydraulics. Note that a perfectly consistent group of experts
would have said that powertrain is three times as important as hydraulics, but in practice they
may not be this consistent. Disparities like this are not unexpected, and are taken into account by
the AHP. The AHP then calculates the column sums for each feature (last row of the “Pairwise”
portion of Table 2), normalizes each column, and takes row averages to determine the final relative
importance for each feature, as shown in the “Normalized” portion of Table 2. A final check is then
performed to make sure that any inconsistencies are within an acceptable tolerance.
10
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Now let us assume that the backhoe hydraulics feature has three options: 4, 5, and 6 function,
assigned scores of 15, 50 and 100 by the experts, respectively. When we rescale these scores to a
scale from 0 to 0.1698 (the relative importance of hydraulics from Table 2), we obtain the final
utilities of these three options: 0.0255, 0.0849, and 0.1698, respectively.
The utility of a complete configuration is estimated as the sum of the utilities of its options.
Although this additive method has been studied and used extensively (Keeney and Raiffa 1976),
other ways of estimating option and configuration utilities are also possible (e.g. conjoint analysis
and its many variations, Hauser and Rao 2004). One advantage of our migration list approach is
that it is independent of the way in which utilities are calculated: The only thing it requires is a
score ranking the relative desirability of each configuration.
To validate the utility values calculated for each of the options – for every BHL model in all
customer segments – we conducted a survey asking actual customers to choose among alternate
configurations. Caterpillar then used t-tests to examine the differences between these surveys and
the utility estimates provided by CAT experts. These differences were not statistically significant.
4.3. Building Migration Lists
Although customers of a given segment tend to behave similarly, they are certainly not identical.
To account for variations within each segment, we modify the migration list procedure in several
ways. First, for each segment we randomly perturb the relative importance (and, consequently,
the option utilities) of randomly selected features. The number of features to perturb is an input
parameter (for CAT, this was around three). Given a perturbation factor t (approximately ten),
the change to a feature’s relative importance is randomly drawn from a uniform distribution over
the interval [−t%, +t%]. CAT also did not want customers to have lists containing configurations
too dissimilar from the one purchased. Therefore, a number called disparity factor (around five)
limits how many options an alternative machine can have that differ from MC . Finally, the model
generates the customer’s reservation price and reservation utility; again, these values are randomly
picked from a predetermined interval around the price and utility of MC .1
We collect the above procedures into a Constraint Programming (CP) model (Marriott and
Stuckey 1998) that finds feasible configurations for LC . This CP model needs to know what constitutes a feasible configuration, i.e. which options are compatible. We use configuration rules to
describe these interdependences. For example, for model 420E, one rule is: if a configuration has
option 9R58666 and either option 2139272 or 2139273, then it cannot have option 9R5321. These
1
Note that randomness in our choice model is restricted to the generation of option utilities and reservation values,
which influence the construction of customer migration lists. Once created, these (fixed) lists serve as input to a
deterministic optimization algorithm. Thus we refer to our model as being “randomized,” as opposed to a random
choice model, which typically has a different meaning.
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
11
rules may involve tens of different options; logical conditions of this form can be easily handled by
CP models. After all feasible configurations are found, those that exceed the generated reservation
price or fall short of the reservation utility are pruned from the customer’s migration list.
Next, configurations are sorted in non-increasing order of total utility and LC is truncated, if
desired, while respecting two conditions. First, if LC is truncated MC must always be retained.
Second, we assume customers place MC first, regardless of MC ’s utility, with a certain probability
(the β factor ; for CAT it was between 0.3 and 0.7). This is an attempt to capture the fact that
some customers are attracted to their MC for reasons we cannot capture with utilities.
Migration across different models is also possible. In this case we apply a set of migration rules
that map a purchased configuration M1 of model m1 (e.g. 416E) to its most likely counterpart
M2 , of a different model m2 (e.g. 420E). Once M2 is known, we generate alternatives as if it
were the customer’s original purchase, and include them (together with alternatives to M1 in m1 )
into LC . Because m2 configurations may have higher utilities, when LC is sorted it may contain
almost no highly ranked m1 configurations. Thus, to capture the fact that customer C originally
preferred an m1 machine, we inflate the utilities of all m1 machines in LC by a preference factor
(between 10 and 20%). As a result, LC ends up with machines of both models, but it does not
allow utilities to overemphasize the attractiveness of m2 machines. According to CAT, the plausible
model migrations are from 416E to 420E and from 430E to 420E.
As was done for option utilities, we also conducted an extensive validation study with CAT
experts to evaluate the quality of our migration lists: Lists were repeatedly generated and examined
to determine whether the substitution patterns indicated were realistic. Throughout this process
the experts provided valuable feedback that helped us fine tune our input parameters. After a few
iterations, CAT experts agreed that our migration lists could be safely used by our optimization
algorithm. Before we present that algorithm, we describe how we estimate a key element of our
objective function, the cost of complexity.
5. Capturing Cost of Complexity
Our next task is to estimate how a line reduction might affect costs. This is a significant challenge,
as product variety affects many functional areas. In most areas the direction of the impact is clear:
material planning costs increase as the number of options to be included increases. In some areas,
however, the impact is not straightforward: sales costs may increase as variety increases because
a large line may overwhelm customers and sales personnel; on the other hand, sales costs may
decrease in variety if it is easier to satisfy a demanding customer. We refer to all costs impacted
by the variety of product offerings, i.e. number of features and options, as the cost of complexity.
12
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Table 3
Main business processes impacted by complexity and the corresponding cost measures. A dagger (†)
indicates alternative cost measures used due to lack of data availability.
Department
Purchasing
Processes impacted
Capital tooling
Supplier operations
Ordering
Forecasting
Quoting and training
Price list creation
Training
Publications
Drawing changes
Customer
acquisition
Marketing
Engineering
Order fulfillment
Product support
Original design & development
Attachment forecasting
Sequencer work
Grief resolution
Dealer solution network - calls
Publications - manuals
Warranty costs
Material planning
Operations
Quality
Inventory management
Schedule volatility
Expedition
Initial process setup
Assembly process
Initial setup
Hot test
Cab test
Measure of complexity
Capital tooling cost
Supplier delivery performance†
Cost of customer acquisition†
Budget expenditure†
Cost of engineering changes
Cost of product & component
Cost of new releases
Headcount cost†
Cost of service calls
Cost of publications
Cost of repairs (first 10 hours)
Cost of repairs (during 11-100 hours))
Cost of repairs (above 101 hours)
Inventory handling cost (prime product)
Inventory handling cost (components)
Headcount cost
Freight cost
Man-hour cost, production planning
Man-hour cost, assembly
Cost of initial setup
Hot-test cost (man-hours)
Cab-test cost (man-hours)
In this section we describe how we built a cost of complexity function for CAT. This includes
both data collection and data analysis to determine and estimate the necessary parameters. The
result of this process is used in our optimization model in Section 6.
5.1. Data collection stage
In conjunction with CAT experts we selected nine functional areas that we felt were most significantly affected by product complexity. We met with these areas’ representatives in a focus group
setting that also included an information systems (IS) representative and a project manager from
CAT. The primary goals of these meetings were: (i) to identify up to three major processes within
each functional area most impacted by product complexity; (ii) to understand which cost-measures
capture the impact of complexity for each major process; and (iii) to identify particular product
features and/or options that have the largest impact on the complexity cost.
We then contacted an accounting representative from CAT who, working with the IS representative and the functional areas, identified which of the identified cost measures were obtainable. For
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
13
some of the processes we identified there was no appropriate accounting data available; hence, we
used an alternative cost measure as a proxy. Table 3 lists the functional areas, processes impacted,
and measures used; when alternate cost measures are used they are denoted by a dagger (†). We
elaborate on several cost measures in Table 3.
Cost of supplier delivery performance refers to a program targeted towards improving availability;
as a part of this program, CAT contacts suppliers with low delivery performance to improve their
processes. We use the cost allocated to this program as a proxy for the cost of supplier operations.
In the customer acquisition department, CAT calculates sales variance cost by tracking all the
discounts that go into making a sale: invoice, extended service, cost of free attachments, etc. We use
this measure to approximate cost of customer acquisition. We rely on CAT’s accounting system for
cost estimates of engineering changes (mainly consisting of payroll to engineers working on changes)
and engineering of new releases (mainly consisting of the payroll of developers and engineers who
work on new parts, and costs of testing and design equipment).
Another outcome of our focus group discussions was the observation that there were two distinct
ways in which complexity affects different business departments: Variety Based and Attribute
Based. We discuss this distinction now.
5.1.1. Variety Based vs. Attribute Based complexity. Certain processes are impacted
by the number of options offered for a feature, while other processes are more impacted by the
presence of specific options. For example, material planners need to calculate requirements for each
type of bucket offered, which increases the departmental cost for material planning. If one type of
bucket is eliminated, the cost of complexity will go down proportionally, regardless of which option
is eliminated. We refer to this effect as Variety Based Complexity, or VBC.
In contrast, for the Cab/Canopy feature, the Deluxe Cab with Air Conditioning is significantly more complicated than a canopy. Hence, if we were to reduce the number of options in the
Cab/Canopy feature, the change in cost would be different depending on which particular option
was eliminated. We refer to this effect as Attribute Based Complexity or ABC. Note that attribute
based complexity is not limited to single options; it is the combination of Deluxe Cab with Air Conditioning that drives complexity cost. In Table 4, we summarize the features that were identified
as most important during our focus group meetings, along with their VBC or ABC classifications.
In addition to the VBC/ABC classification, there were two other important considerations that
came out of our focus group discussions: time lag and volume drivers of complexity.
5.1.2. Time lag of the impact of complexity Our discussions revealed that the effect of
complexity on different processes may be felt at different times. For example, assembly cost today
is impacted by the product complexity being built today, while warranty costs are affected by
14
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Table 5
VBC
VBC
VBC
ABC VBC
VBC VBC
VBC
VBC VBC
VBC VBC
VBC
VBC
VBC
VBC
VBC
VBC
VBC VBC VBC
VBC
ABC
Sticks
Cab/Canopy
Control Groups
Hoe Buckets
Quick Coupler
Auxiliary Lines
Loader Buckets
Department/Feature
Purchasing
Customer acquisition
Marketing
Engineering
Order fulfillment
Product support
Material planning
Operations
Quality
ABC/VBC classification of features.
Backhoe Hydraulics
Table 4
VBC VBC
VBC
VBC
ABC
VBC VBC
VBC
VBC
ABC ABC
ABC
Summary of cost time lags (in months).
Department
Time Lag
Purchasing
0
Customer acquisition
3
Marketing
0
Engineering:
Changes
-6
Releases
-8
Product and component costing
-7
Order fulfillment
0
Operations
0
Quality
0
Department
Time Lag
Product support:
Service calls
6
Repairs in the first 10 hours
4
Repairs in 10-100 hours
9
Repairs after 100 hours
9
Material planning:
Prime product inventory
2
Scrap of surplus materials
6
Inventory scrap
6
complexity that was offered a certain time ago (positive time lag), and engineering and marketing
costs may be impacted by the complexity that will be offered in the future (negative time lag).
Based on our focus group discussions we estimated the number of months of time lag for different
cost pools. Later (in Section 5.2.1), we use this information in estimating the parameters of the
cost of complexity function. Table 5 summarizes our analysis of time lag parameters.
5.1.3. Volume drivers for complexity Another important aspect in understanding the
impact of complexity is understanding different volume metrics and how they affect costs. Not
surprisingly, most areas are affected by sales volume: costs increase as more items are produced and
sold. However, costs are also driven by other volumes. For example, product support is impacted
by the number of unique configurations built, because quality may decrease when employees have
to work on many different configurations. Other processes, such as assembly planning and engineering, are impacted by the complexity offered, as CAT should be ready to assemble any feasible
configuration offered. Later (again, in Section 5.2.1), we use this information to determine which
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Table 6
15
Volume drivers.
Department/Volume driver Sales volume Complexity offered Complexity built
Purchasing
X
X
Customer acquisition
X
X
Marketing
X
X
Engineering
X
Order fulfillment
X
X
Product support
X
X
Material planning
X
X
X
Operations
X
X
Quality
X
X
predictors should be included in the cost model for each department. Table 6 provides a summary
of the most important volume drivers for each department.
5.2. Cost of complexity function
Before defining the cost functions and notation associated specifically with VBC and ABC costs, we
present supplementary notation that will be common to both. We use small letters for superscripts
and subscripts, bold letters for sets, and capital letters for numbers coming from collected data.
Supplementary Notation:
d∈D
Superscript used to represent attributes pertaining to department d,
where D is the set of all departments considered in the study;
f ∈F
Superscript used to represent attributes pertaining to feature f ,
where F is the set of all features in the product line;
d
F ⊂ F The set of all features identified as relevant for department d;
o∈O
Superscript used to represent attributes pertaining to option o,
where O is the set of all options in the product line;
Of ⊂ O The set of all options in feature f ;
Nf
Number of options in feature f : |Of |;
Nd
Set of cardinalities of all features relevant for department d: N d = {Nf : f ∈ F d }.
5.2.1. Estimation of VBC effect To estimate the VBC effect for each department d, CVd BC ,
we use the Cobb-Douglas log-linear function. The Cobb-Douglas function is frequently used for
estimating non-linear relationships (see Greene 2000); it can capture different returns to scale and
has several attractive analytical properties. Two properties are of particular convenience for us:
First, the log transformation of the Cobb-Douglas function is linear and hence can be estimated
using linear regression. Second, a partial derivative of the Cobb-Douglas function has a simple form
that is useful in the derivation of the differential cost of complexity (Section 5.3). After estimating
the models, we validate our Cobb-Douglas function via several goodness of fit tests (e.g. Figure 2).
Using lower-case Greek letters to represent estimated parameters, the Cobb-Douglas function
for CVd BC is given by
16
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
d
CVd BC (N d , V, U ) = ξ d V α U β
d
Y
f ∈F
Where:
V
U
ξd
αd
βd
γfd
γd
Nf f .
(1)
d
Total sales volume;
Total number of unique configurations sold;
The size of the cost pool at department d relative to other departments;
The effect of the sales volume on the cost of complexity;
The effect of the number of unique configurations sold on the cost of complexity;
The effect of the cardinality of feature f on the cost of complexity;
We fit this function to the data to estimate ξ d , αd , β d , and γfd for all d and f . Using data collected
for 60 months, from January 2001 to December 2005, for all cost measures summarized in the third
column of Table 3. We also collected data from the price lists from 2000 to 2006 to capture all
changes in the option offerings, which were used as independent variables. (Due to the time lags
identified in Section 5.1.2, we collected price list data over a wider time range than cost data.)
Similarly, we collected monthly sales (V ) and number of unique configurations sold per month (U )
from 2000 to 2006 and ran regression analysis to confirm our hypotheses concerning lags in Table 5.
The nature of the data suggests that there is serial correlation, which was confirmed with the
Durbin-Watson statistic for all cost pools. Therefore, we use the Yule-Walker approach to fit
the data to the log transformation of CVd BC (N d , V, U ) (Greene 2000). The log transformation of
CVd BC (N d , V, U ) used for estimation, with as the model error term, is
Log[CVd BC (N d , V, U )] = Log[ξ d ] + αd Log[V ] + β d Log[U ] +
X
f ∈F
γfd Log[Nf ] + .
(2)
d
Next, we fitted collected data to equation (2) and obtained statistical models for all departments.
We analyzed our model using both graphical and numerical tests: we examined the plots of residuals
for normality, heteroscedasticity, and for influential outliers (an example of these plots for one of
the cost pools, customer acquisition cost, is depicted in Figure 2). We also checked for significance
of the models, and only accepted models with reasonable coefficients of variation (as summarized
in Table 7) and low RMSE. The estimates that are statistically significant at the p = 0.05 level are
marked with an asterisk in Table 7. Hence, not all of the originally identified departments were
included in the final cost of complexity function, and some of the sets F d were reduced.
First, we comment on the coefficient of determination of the models (R2 ) in Table 7. Some of
the departmental costs are heavily impacted by factors outside CAT’s walls: for example, cost
of customer acquisition is impacted by competitors’ actions, and cost of supplier delivery performance is impacted by the suppliers’ operations. Hence, we expect the coefficients of determination
to be lower for such departments. The results in Table 7 are consistent with our expectations.
The SAS System
17:05 Tuesday, October 24, 2006 3
The REG Procedure
Model: linear
ariable: dealer_trades
dealer_trades
Figure 2
Goodness of fit plots for the customer acquisition cost model.
17
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
The SAS System
The REG Procedure
Model: MODEL1
Dependent Variable: DRF1_l
(a)
Normal Q-Q plot
(c)
Cook’s distance
(b)
Histogram of residuals
Table 7
Fn.
CA
MP
MP
ENG
ENG
PS
PS
PS
P
(d)
Residuals vs. predicted values
Fit results. An asterisk * indicates statistically significant parameters at p = 0.05 level. Subscripts HC, C,
CW and H stand for hydraulic combinations, cabs, counterweights, and hydraulics.
Cost pool (d)
Customer acquisition
Inventory (components)
Inventory (prime product)
Engineering changes
Engineering product & component
Repair costs in the first 10 hours
Repair costs during 11–100 hours
Repair costs above 101 hours
Supplier delivery performance
R2
0.37
0.29
0.38
0.41
0.55
0.29
0.32
0.55
0.33
ξd
962771120.4*
1087617.41*
297.68
0.00342591
6.39E-04*
26238.94*
3055.00*
934.63*
36.65
αd
0.19
−0.03
0.91*
−0.21
−0.32
−1.41*
βd
d
γHC
d
γC
−1.13*
1.04*
4.81712*
4.107*
−0.54
d
γCW
d
γH
−1.68*
0.91
3.14*
0.59*
1.28*
2.78*
1.5*
Similarly, when analyzing the models for warranty costs, we expect that earlier failures would be
less predictable, due to learning effects. The results are again consistent with this expectation; the
coefficient of determination increases for repair costs as the time of failure increases. Nevertheless, in order to ensure that our results are robust, we use ranges of parameters in running the
optimization model as described in Section 7.
We comment on some of the findings from Table 7. First, increasing the number of options
CA
for hydraulics and cabs decreases the cost of customer acquisition: i.e. γH
< 0 and γCCA < 0. A
large proportion of the cost of customer acquisition consists of sales variance or discounts given
18
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Figure 3
Actuals versus 95% confidence interval of predictions for customer acquisition.
to customers by dealers in order to attract business. Cabs and hydraulics are very important
considerations for customers; having a large selection of options for these features makes it easier
for the dealer to make a sale, decreasing the sales variance. This finding is in line with the intuition
of sales and marketing representatives from CAT. However, this was the first time that CAT was
able to quantify this effect.
Another interesting observation is that sales volume has a negative impact on product support
costs (i.e. repair costs). When volume goes up CAT employees assemble more machines with the
same options, employees get more experienced with particular options, and the number of mistakes
decreases. This intuition again seemed plausible to CAT, but had never been quantified. Guided by
our study, CAT subsequently has performed similar cost of complexity analysis in other product
divisions and obtained comparable results.
Finally, we tested our predictions by using the first year of data to fit the model, then calculated
the predicted values for the next four years, and finally compared them to the actual data. Figure 3
shows that a large majority of the actual values lie within the 95% confidence interval around the
predicted values, again for customer acquisition cost.
5.2.2. Estimating ABC effect From focus groups, we identified that ABC effects were observed primarily in three departments: assembly, production planning, and engineering. The nature
of work in these departments suggested that the relationship between ABC cost and complexity
is linear: if it takes 5 extra minutes to install a particular option on a machine, this cost will
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Table 8
19
ABC costs.
Option
δo
ωo
Any option
$1,000.00
In addition:
IT
$6,000.00 $23.61
Cab
$14,656.25 $53.80
E-Stick
$125.00 $1.30
One Way Line $3,593.75 $5.64
Ride Control
$1,281.25 $5.86
apply to each machine that contains the option. This coincided with expert opinion, which posited
that learning effects are minimal. Hence we model the total cost of offering and producing option
o
o, CABC
, as the sum of the one-time cost across all departments if option o is offered and the
incremental cost across all departments that is incurred each time a machine with option o is built:
o
CABC
(ao , V o ) = δ o ao + ω o V o .
Where:
δo
ao
ωo
Vo
V
Cost incurred if option o is offered;
Binary variable that indicates whether option o is offered or not;
Cost incurred each time option o is produced;
Number of configurations sold that contain option o;
Set of number of configurations sold for all options: V = {V o : o ∈ O }.
The cost parameters δ o and ω o were estimated using expert opinions, time studies, and accounting
information from all functional areas that identified this option as important; see Table 8.
5.3. Differential cost of complexity function
Instead of computing the total revenue and total cost of each line, our complexity cost function
enables us to start with the current line and compute the estimated change in revenue and the
estimated change in cost as the number of features and options change. To compute this “differential” cost of complexity, we take the partial derivatives of CVd BC (N d , V, U ) with respect to all
variables, and then combine them into a differential cost of complexity function. Specifically, the
change in cost of complexity when the number of options for feature f changes is:
γfd d αd β d Y γid
γfd d
dCVd BC (N d , V, U )
=
ξ V U
Ni =
CV BC (N d , V, U ),
dNf
Nf
N
f
d
(3)
i∈F
substituting the definition of
CVd BC (N d , V, U )
given by (1).
This differential cost of complexity function includes the predicted value of CVd BC (N d , V, U ), the
average size of the cost pool at department d. This is problematic for two reasons: (i) The predicted
CVd BC (N d , V, U ) contains an error term and hence may not give an accurate size of the cost pool;
20
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
and (ii) Values of V , Nf , and U change during the specified time period. Therefore we approximate
CVd BC (N d , V, U ) with the historical average cost at department d (Dd ) attributable to complexity,
taken over the period of 60 months from January 2001 to December 2005:
γfd d
dCVd BC (N d , V, U )
≈
D .
dNf
Nf
(4)
Using (4) the total change in complexity cost due to a change in the number of options offered
for feature f is then estimated as:
Dd
γfd
∆Nf ,
Nf
(5)
where ∆Nf represents the change in the number of options in feature f after line optimization;
similarly we will precede V , U , V o , N , N d , and V with ∆ to represent change. We will formally
define these functions in Section 6 when we introduce the optimization model.
In an analogous fashion we differentiate the cost of complexity function with respect to volume
and number of unique configurations. This yields the total differential cost of complexity DVd BC :
DVd BC (N d , V, U, ∆N d , ∆V, ∆U ) =
X
f ∈F
Dd
d
γfd
βd
αd
∆Nf + Dd ∆U + Dd ∆V.
Nf
U
V
(6)
To capture the differential ABC effect, we must account for the change in the number of options
offered and sold. If we eliminate an option o from the price list, the cost of complexity will decrease
by δ o and if sales with option o deviate from V o , the cost will change accordingly:
DABC (∆V , a) =
X
δ o (ao − 1) +
o∈O
X
ω o ∆V o , where a = {ao : o ∈ O }
(7)
o∈O
Finally, the total differential cost of complexity is:
D(N d , V, U, ∆N d , ∆V, ∆U, ∆V , a) =
X
DVd BC (N d , V, U, ∆N d , ∆V, ∆U ) + DABC (∆V , a). (8)
d∈D
Because the complexity cost function is non-linear, this approach is accurate only for small
changes. The accuracy of this approximation may be tested by calculating the actual change in
total cost of complexity CV BC between the original product line and the optimized product line, and
then comparing this to the result of the approximation DV BC . In our experiments, the differential
cost approximation was within 5% of the calculated change in the cost of complexity, which was
acceptable for the purposes of our project2 . As a result, equation (8), with estimated parameters,
becomes a part of the objective function in the optimization model of Section 6.
2
If desired, this approximation may be improved by using an iterative approach built into the optimization model,
in which the differential is re-estimated after each small change in variables.
21
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
6. The Optimization Model
With the migration lists described in Section 4, which contain the customer flexibility information,
and equation (8) of Section 5, which contains the differential complexity cost information, we can
now describe our optimization model: We use a mixed-integer program to select the set of options
and configurations offered, and their prices, to maximize total revenue from sales minus complexity
costs. We use customer purchases in 2006 as the basis for generating our migration lists.
In addition to the data defined in Section 5, our optimization uses the following data:
• I — The set of all customers;
• Li — The ordered set of configurations in the migration list of customer i ∈ I;
• J — The set of all possible configurations;
• Oj — The set of all options in configuration j ∈ J ;
• Ri — Reservation price of customer i ∈ I; this could be made machine dependent if desired;
• M — Maximum reservation price over all customers (M = maxi∈I Ri );
• Cj , Cj0 , Bj , Pj — Cost, variable cost, base price, and current sale price of configuration j ∈ J ,
respectively. The variable cost Cj0 is an internal CAT figure, lower than Cj , the purpose of which
is to enforce a constraint on the average profit margin over all machines sold (see (16)).
The decision variables are as follows (ao was already defined in Section 5.2.2):
• ao = 1 if option o ∈ O is available, 0 otherwise.
• qj = 1 if configuration j ∈ J is bought by some customer, 0 otherwise;
• xij = 1 if customer i buys configuration j, 0 otherwise (i ∈ I, j ∈ Li );
• po — Price of option o ∈ O (po ≥ δ o , where δ o is defined in Section 5.2.2);
• rij — Profit obtained from customer i if she purchases configuration j (i ∈ I, j ∈ Li ).
We now provide precise definitions to the following terms that appear in (6) and (7):
∆Nf =
X
∆U =
X
ao − Nf ,
o∈Of
qj − U,
j∈J
∆V =
XX
xij − V,
i∈I j∈Li
∆V o =
X
X
xij − V o .
i∈I j∈Li |o∈Oj
The objective function maximizes the total profit from sales (
P P
rij ) minus the differential
i∈I j∈Li
cost of complexity given by equation (8) of Section 5:
max
XX
i∈I j∈Li
rij −
X
d∈D
DVd BC (N d , V, U, ∆N d , ∆V, ∆U ) − DABC (∆V , a) .
22
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Our optimization model uses the following constraints, where max inc is the maximum allowed
percentage price increase for a configuration:
qj ≤ ao ,
∀ j ∈ J , o ∈ Oj
xij ≤ qj ,
X
(9)
∀ i ∈ I, j ∈ Li (10)
xik + qj ≤ 1,
∀ i ∈ I, j ∈ Li (11)
k after j in Li
rij ≤ Bj +
X
po − Cj ,
∀ i ∈ I, j ∈ Li (12)
o∈Oj
Bj +
X
rij ≤ (min{Ri , Pj (1+max inc)} − Cj )xij
∀ i ∈ I, j ∈ Li (13)
po ≤ Ri + (Pj (1+max inc) − Ri )(1 − xij )
∀ i ∈ I, j ∈ Li (14)
o∈Oj
Cj (1 + min marg) ≤ Bj +
X
po ≤ Pj (1 + max inc),
∀j∈J
(15)
o∈Oj
min avgmarg
XX
(rij + Cj qj ) ≤
i∈I j∈Li
XX
(rij + (Cj − Cj0 )qj ).
(16)
i∈I j∈Li
Constraint (9) says if option o is not available (ao = 0), no configuration that contains it can be
bought (qj = 0); (10): if configuration j is not bought (qj = 0), make xij = 0; (11): if j was bought
by someone (qj = 1) it must be available; therefore customer i will not consider configurations with
lower priority (k after j in Li ); (12): profit is at most equal to price minus cost; (13): profit is limited
by the customer’s reservation price Ri , and no purchase (xij = 0) means no profit (rij = 0); (14): if
customer i bought configuration j (xij = 1), price of j must be no more than Ri ; (15): configuration
margins have to be at least min marg and configuration prices cannot increase more than max inc
above original values; (16): average margin over all machines has to be at least min avgmarg.
We can turn off the price optimization aspect of the model by removing variables po , constraints
P
(14)–(16), making max inc = 0 in (13), and substituting Pj for Bj + o∈Oj po in (12). (This
additive price structure is in line with CAT’s pricing rules.)
Our optimization models have around 850,000 variables and 1.8 million constraints, and they
are solved using ILOG CPLEX Optimizer with default parameters. Typical solution times range
from 6 to 8 hours, including preprocessing.
7. Results from Our Analysis
CAT’s goal was to make a drastic reduction in the number of configurations offered without significantly impacting customer satisfaction and market share. How to achieve such a goal, or whether
such an outcome was even possible, was unclear at the outset of the project: Optimizing their
product line required careful understanding and modeling of potential losses due to reduced offerings, and potential savings due to reduced cost of complexity. Since this reduction would present
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
23
customers with fewer configurations, CAT assumed that each remaining configuration could be
priced a little lower; the reduced cost of complexity would allow this while maintaining profit.
Throughout our analysis, to ensure that our recommendations were robust we ran the optimization model across a range of parameters and migration list lengths.
7.1. Stage 1: Focusing on Configuration and Option Reduction
As an initial benchmark for our optimization, we first sought to identify the set of configurations
that maximized profit at the current option prices, assuming limited customer migration (no more
than a dozen configurations on a migration list, and no migration across models). While profits did
increase, we obtained very little reduction in the number of configurations, except when a reduction
P
was explicitly enforced by the constraint j∈J qj ≤ (1 − min conf red)|J |. Moreover, forcing a
large reduction resulted in a significant decrease of sales revenue.
Upon reflecting on the results, we hypothesized that further reducing the cost of complexity
would require significant cuts in options. Hence, a new constraint was added to the model to
P
force option reduction: o∈O ao ≤ (1 − min opt red)|O |. We then re-ran the optimization model
forcing a reduction in the number of configurations and the number of options. Reducing options
did succeed in reducing configurations, in some solutions by as much as 94%, and increased profit
by generating a large cost of complexity reduction. But it also resulted in a drop in sales volume
of up to 67%. This disturbed CAT team members since the company has always prided itself on
its market share. Fulfilling customer demand therefore became an important new metric of the
analysis.
7.2. Stage 2: Opening Up Choices
Thus in the next phase of our analysis we wanted to explore what results would be possible if
customers were significantly more flexible, possibly as a result of price incentives. We modeled
this flexibility in two ways: We increased the migration list length (to 100 configurations) and we
enabled model-to-model migration.
This approach started to generate encouraging results. A solution emerged with an increase in
profit of 8.8%, less than a 2% reduction in sales volume, and a reduction in configurations equal
to 65%. But further analysis showed that the number of options had not decreased significantly.
The increase in profit came from a decrease in the number of configurations, and increases in price
paid by customers who migrated to slightly more expensive configurations. Performing sensitivity
analysis confirmed this conclusion: expecting a large reduction in options was not realistic. However,
a large reduction in configurations was possible.
24
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Figure 4
Final product hierarchy and packages for the BHL 420E series.
This resulted in a problem for CAT: Without a reduction of options, how would this new set of
configurations be presented to the customers? Restaurants can get by with a 3-4 page menu that
lists all their entrees. But no customer would flip through a menu of 70-90 pages listing all the
possible BHL configurations. A new scheme had to be devised.
7.3. Stage 3: Standardization and Options Packages
We decided to try two new strategies to concentrate customer demand on a manageable number of
configurations. The first strategy was standardization: Could options such as High Ambient Cooler
and Engine Heater be made standard across all configurations? Optimization models with these
options forced into every configuration yielded cost reductions that justified a reduction in price
large enough to make the standardized configurations attractive to customers, while maintaining
sales volumes and profit. Other rarely used options were eliminated using a similar approach. For
example, the Cab/Canopy options were cut from five to two.
The second strategy was creating packages of options commonly found together. For example,
guided by customer segment preferences, a single pair of loader hydraulics and powertrain options
most likely to meet each segment’s needs was proposed. Manual inspection and cluster analysis of
the best solutions found so far led to the discovery of other options often found together.
The optimization was then run assuming standardization and option packaging, with constraints
on the maximum price increase. This yielded the final product hierarchy for the 420E series,
shown in Figure 4: It consists of 9 base-machine-assembly (BMA) packages, 5 finished-to-order
(FTO) packages, and 3 hydraulics options, for a total of 135 possible configurations, some of them
anticipated to be much more popular than others.
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
25
Pricing optimization showed that with these 135 configurations, revenue from sales could increase
by almost 7% and profit by 15%, with 99.6% demand fulfillment. Interestingly, 76% of the projected
cost of complexity savings came from reductions in finished goods inventory and warranty costs
(in particular, the cost of addressing failures in the first 100 hours of machine operation). Since the
goal of the project was to maintain similar profit levels (rather than seeking increases), the team
determined appropriate option price reductions to drive dealer behavior toward these configurations
while maintaining profit. The proposed pricing policy resulted in an anticipated reduction in profit
from sales of 4%, which was easily made up by the reduction in cost of complexity to yield a
total profit increase of 4.8%. We had finally found the very small subset of configurations that we
believed was broad enough to satisfy CAT’s customers and dealers, but also focused enough to
drive operational and supply chain efficiencies.
This final recommendation was presented to CAT, and approved.
8. Implementation Details
8.1. Initial Implementation
The implementation of such a dramatic change faced challenges. The first challenge was a concern
that our solution would restrict customer choice too much, resulting in excessive lost sales. Caterpillar therefore showed the proposed packages to a subset of dealers, including their top 15 dealers,
responsible for nearly half of the North American demand. The dealers suggested configurations
to add and remove, and we adapted some of the packages slightly. Caterpillar also suggested the
addition of high-cost options, but the dealers did not support this, so they were excluded.
Overall, the response from the dealers was very favorable, with many dealers projecting they
would be able to satisfy 80% of their demand from these packages. In contrast, dealers estimated
that the machines they currently kept in stock satisfied only 5-10% of their demand. Thus, standardization and option packaging represented an effective way of potentially reducing configurations.
The second challenge came with the recession of 2009. Suddenly, every sale became crucial,
and configuration reduction became a lower priority. However, with the economy showing signs of
recovery in 2010 and with some changes to product sourcing at Caterpillar, reducing the number
of configurations again gained importance. Caterpillar therefore put an updated price list for the
430E product line into effect in April 2010; but, in order to minimize the risk of lost sales in the
still difficult economic environment, the new price list was introduced alongside the old price list,
rather than as a replacement.
Concurrent with this new price list, now featuring 124 configurations, Caterpillar introduced a
3-lane strategy for order fulfillment:
26
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
• Lane 1, the fastest lane, projected that orders on the four designated “most popular” fully
configured machines would be satisfied within a few days.
• Lane 2 projected that orders on the remaining 120 choices, broken into the Loader, Comfort
and Convenience, and Excavation packages, would be satisfied within a few weeks.
• Lane 3 projected any a-la-carte machine would be satisfied within a few months, as previously.
Upon implementation, Caterpillar experienced positive dealer feedback and large reductions
in the number of unique configurations sold. This contrasted with previous attempts to reduce
the size of the product line, based on a Pareto analysis of the top few dealers. These had likely
been ineffective, in CAT’s opinion, because they did not effectively model the interaction of cost,
customer preferences, and substitution, as we had.
It is worth noting that, to the best of our knowledge, neither standardization nor option packaging
were being considered prior to the start of our project. These concepts were only developed after
our analysis indicated that CAT needed to find a way to limit the number of configurations while
not eliminating too many options. Our analysis then gave CAT direction in implementing these
new strategies, by indicating which sets of configurations were likely to be successful.
8.2. Moving Forward
In 2011 Caterpillar consolidated the two price lists into a single price list featuring options packages.
They were able to capture over a quarter of their sales volume just in Lane 1; in contrast, the
four top-selling configurations Caterpillar used to offer before this project captured 11.3% of both
sales and total revenue for the 420E model (see Figure 1). In addition, CAT enjoyed a reduction
in warranty costs, attributable to many factors – including this project. Caterpillar has continued
to focus their BHL offerings, for example reducing the 420E (now the 420F) Lane 1 offerings to
three base machines by the middle of 2014. The other BHL lines have seen similar reductions.
Caterpillar has rolled out the lane approach to the entire company, making it an integral part
of CAT’s business strategy (Thomson Reuters (2014)). While the methodologies used to determine the lane offerings in other divisions were somewhat simpler than the analysis done here, our
work provided support for the corporate-wide Lane Strategy. In particular, our cost of complexity
analysis approach has been applied to other divisions, such as Wheel Loaders, with the goal of
capturing all the benefits and understanding all the consequences of proposed line changes. The
detailed analysis and structured optimization approach reported in this paper allowed Caterpillar
to counteract skepticism toward the lane approach embarked upon by BCP, prior to the CEO
mandate, and successful implementation, of this strategy for the entire company.
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
27
9. Conclusion
We present a three-step procedure to restructure a product line, and demonstrate its successful
application on the Backhoe Loader line at Caterpillar. Our methodology hinges on (i) The construction of migration lists to capture customer preferences and willingness to substitute; (ii) Explicitly
capturing the (positive and negative) cost of complexity of offering a specific product line across
different functional areas; and (iii) integrating these tools within a mathematical programming
framework to produce a final product line. One of the greatest strengths of our methodology is its
flexibility – each step can be custom tailored to a company’s particular setting, data availability
and strategic needs, so long as it produces the necessary output for the next step of the algorithm.
As Caterpillar evolves, so has their Lane Strategy. For example Thomson Reuters (2014) discusses
a new variation in which lanes may contain only partially completed machines, which can be finished
to customer order as needed. Ideas such as these offer opportunities to extend our work to new
problem domains – analytically characterizing the performance of such a delayed differentiation
strategy is an exciting and challenging problem.
Outside of Caterpillar, there are several directions that our work can be extended. First, explicit
experimental validation of our empirical models – in particular our migration approach – would be
of value. While this approach has been successfully applied in the construction equipment industry,
further study could help establish how it could be applied in other sectors, and what changes
might be necessary. For example, we set the list lengths based on consultation with Caterpillar
executives. A better understanding of how long such lists really are, and how willing customers are
to substitute (and whether there is any sort of explicit cost to this) would be of interest.
Second, as our methodology is applied to other settings, new constraints might need to be
incorporated into our mathematical program. One of the benefits of our procedure is that our math
programming formulation is flexible enough to accommodate other such constraints. Nevertheless,
how it performs in other settings needs to be established.
Finally, the central thrust of this paper has considered the trade-off between cost of complexity
and product line breadth. Caterpillar has found what they believe to be the correct trade-off, which
entailed a dramatic reduction in their product line. Different companies, in different industries,
will have to answer this question for themselves. It is our hope that the methods we present in our
paper can likewise help them find the answers they seek.
References
Alptekinoglu, A., C. Corbett. 2008. Mass customization vs. mass production: Variety and price competition.
Manufacturing and Service Operations Management 10(2) 204–217.
28
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
Anderson, S. 1995. Measuring the impact of product mix heterogeneity on manufacturing overhead cost.
The Accounting Review 70(3) 363–387.
Banker, R., G. Potter, R. Schroederr. 1995. An empirical analysis of manufacturing overhead cost drivers.
Journal of Accounting and Economics 19 115–137.
Belloni, A., R. Freund, M. Selove, D. Simester. 2008. Optimizing product line designs: Efficient methods
and comparisons. Management Science 54(9) 1544–1552.
Bitran, G., J.-C. Ferrer. 2007. On pricing and composition of bundles. Production and Operations Management 16(1) 93–108.
Broniarczyk, S., W. Hoyer, L. McAlister. 1998. Consumers’ perceptions of the assortment offered in a grocery
category: The impact of item reduction. Journal of Marketing Research 35 166–176.
Chen, K., W. Hausman. 2000. Technical note: Mathematical properties of the optimal product line selection
problem using choice-based conjoint analysis. Management Science 46(2) 327–332.
Chen, Y., J. Carillo, A. Vakharia, P. Sin. 2010. Fusion product planning: A market offering perspective.
Decision Sciences 41(2) 325–353.
Chen, Y., A. Vakharia, A. Alptekinoglu. 2008. Product portfolio strategies: The case of multifunction
products. Production and Operations Management 17(6) 587–598.
Chong, J., T. Ho, C. S. Tang. 1998. Product structure, brand width, and brand share. T. Ho, C. S. Tang,
eds., Product Variety Management: Research Advances. Kluwer Academic Publishers, 39–64.
Fisher, M., C. Ittner. 1999. The impact of product variety on automobile assembly operations: Empirical
evidence and simulation analysis. Management Science 45(6) 771–786.
Fisher, M., R. Vaidyanathan. 2011. An algorithm and demand estimation procedure for retail assortment
optimization. Working Paper, University of Pennsylvania .
Foster, G., M. Gupta. 1990. Manufacturing overhead cost driver analysis. Journal of Accounting and
Economics 12 309–337.
Fruchter, G., A. Fligler, R. Winer. 2006. Optimal product line design: Genetic algorithm approach to mitigate
cannibalization. Journal of Optimization Theory and Applications 131(2) 227–244.
Greene, W. H. 2000. Econometric Analysis. 4th ed. Prentice Hall, New Jersey.
Hauser, J. R., V. Rao. 2004. Conjoint analysis, related modeling, and applications. J. Wind, P. Green,
eds., Advances in Market Research and Modeling: Progress and Prospects. Kluwer Academic Publishers,
Boston, MA, 141–168.
Keeney, R. L., H. Raiffa. 1976. Decisions with Multiple Objectives: Preferences and Value Trade-offs. Wiley,
New York.
Kok, G., M. Fisher. 2007. Demand estimation and assortment optimization under substitution: Methodology
and applications. Operations Research 55(6) 1001–1021.
Shunko et al.: Restructuring the Backhoe Loader Product Line at Caterpillar
29
Kok, G., M. Fisher, R. Vaidyanathan. 2009. Assortment planning: Review of literature and industry practice. N. Agrawal, S. Smith, eds., Retail Supply Chain Management. Kluwer Academic Publishers,
Amsterdam, The Netherlands, 1–55.
Marriott, K., P. Stuckey. 1998. Programming with Constraints: An Introduction. MIT Press.
Rash, E., K. Kempf. 2012. Product line design and scheduling at Intel. Interfaces 42(5) 425–436.
Saaty, T. L. 1980. The Analytic Hierarchy Process. McGraw-Hill Book Co., New York.
Schoen, C. 2010. On the optimal product line selection problem with price discrimination. Management
Science 56(5) 896–902.
Tang, C. 2010. A review of marketing-operations interface models: From co-existence to coordination and
collaboration. International Journal of Production Economics 125(1) 22–40.
Tang, C., R. Yin. 2010. The implications of costs, capacity, and competition on product line selection.
European Journal of Operational Research 200(2) 439–450.
Thomson Reuters. 2014.
script.
Caterpillar Inc. 2014 Analyst Meeting at ConExpo, edited tran-
Accessed on 8/19/2014.
http://phx.corporate-ir.net/External.File?t=1&item=
VHlwZT0yfFBhcmVudElEPTUwODQxOTB8Q2hpbGRJRD01MzYwODg=.
Wang, X., J. Camm, D. Curry. 2009. A branch-and-price approach to the share-of-choice product line design
problem managing product variety. Management Science 55(10) 1718–1728.
Ward, Julie, Bin Zhang, Shailendra Jain, Chris Fry, Thomas Olavson, Holger Mishal, Jason Amaral, Dirk
Beyer, Ann Brecht, Brian Cargille, Russ Chadinha, Kathy Chou, Gavin DeNyse, Qi Feng, Cookie
Padovani, Sesh Raj, Kurt Sunderbruch, Robert Tarjan, Krishna Venkatraman, Joseph Woods, Jing
Zhou. 2010. HP transforms product portfolio management with operations research. Interfaces 40(1)
17–32.
Yunes, T. H., D. Napolitano, A. Scheller-Wolf, S. Tayur. 2007. Building efficient product portfolios at John
Deere and Company. Operations Research 55(4) 615–629.